Our new mattress testbed scored 97 on its own SEO checker. Google barely surfaced it. ChatGPT still cited it as "Best for Singapore" for one query.

All three observations were true. Together, they showed that our score was answering the wrong question.

Ninety-seven meant eligible, not trusted

The checker was good at page hygiene: titles, schema, internal links and technical structure. A competitor scored 87 on the same rubric. Across 45 Singapore SME sites, 42 responded and the median score was 92.

That made 97 look impressive. It did not create backlinks, brand demand or search history.

Rankr benchmark comparing search readiness across six Singapore SME sectors
The benchmark showed that strong on-page scores were common and did not prove real search visibility.

The ChatGPT result was interesting for a different reason. The page carried concrete, extractable facts about Singapore weather, room size and mattress fit. Those facts matched one answer well enough to be retrieved despite the weak domain authority.

One answer is not a ranking system. It is one useful observation from one query.

The score had to become a measurement stack

We split the original number into four questions:

  1. Can a crawler understand the page?
  2. Does the domain have enough authority to rank?
  3. Are impressions turning into clicks and clients?
  4. Can an AI system extract a useful answer and cite it?
Lullaflex bedroom planner with Singapore room presets and mattress dimensions
The testbed paired useful local facts with a real planning tool instead of publishing another generic mattress page.

The practical result changed how Rankr works. The agent can still fix eligibility, but it no longer pretends that perfect metadata manufactures trust. The interesting opportunity is helping small businesses publish facts that both people and retrieval systems can use, then measuring whether that attention turns into business.